Abstract:

We propose a statistical method for modeling and synthesizing sounds with both sinusoidal and attack transient components. In addition, the sinusoidal component can have pitch-changing characteristics. The method applies multivariate decomposition techniques (such as independent component analysis and principal component analysis) to learn the intrinsic structures that characterize the sound samples. Afterwards these structures are used to synthesize new sounds which can be drawn from the distribution of the real original sound samples. Here we apply the method to impact sounds and show that the method is able to generate new samples that have the characteristic attack transient of impact sounds.